dolfin team mailing list archive

[Benjamin Kehlet <benjamik@xxxxxxxxxxxxxxxxxxxxx>]

Johan Hake wrote:
> On Monday August 23 2010 15:59:15 Benjamin Kehlet wrote:
> > Hello!
> >
> > I have been trying to get the python wrappes of the ODE solvers in place
> > again and learn more about Swig. The wrappers have been broken since the
> > change from uBLASVector to real*.
> >
> > Dolfin needs to call functions implemented on the python side, so
> > directors are needed. This simple typemap seems to do the job:
> >
> > %typemap(directorin) double* ODE_ARGIN
> > {
> >    // Get the length of the arrays
> >    int n = size();
> >    $input = PyArray_SimpleNewFromData(1, &n, NPY_DOUBLE, $1_name);
> > }
> >
> > Some questions from a Swig newbie: Is this the preferred way of doing this?
> > Note that the typemap have to call size() (a member function of class
> > ODE) to get the length of the array, so I guess it is not possible to
> > use a more generic typemap.
>
> This is why we instroduced the dolfin::Array class. So speciallized typemaps 
> like this could be avoided. 
>
> Is there any way you can port the ODE interface to using dolfin::Array 
> instead? Then most of the typemaps should be inplace already.

That shouldn't be too complicated. Or was there a reason for changing to 
plain c arrays (and not dolfin::Array) in the first place? Anders?

> > It seems to me that no "directorargout" typemap is needed when returning
> > arrays, since the numpy arrays are just wrapping the plain c array. Is
> > this safe?
>
> Yes I think so.
>
> > How should "const double*" be handled (I used a const_cast<double *>
> > when I tested it, to avoid compiler errors)? Is data copying needed here?
>
> Should not be any problem as Python does not know what const is.
>
> Johan